Introduction to Ecological Forecasting

Jasper Slingsby

What do you consider when making a decision?

Informing decisions requires knowing (or guessing at) something about the future.

We base our expectation on:

  • The evidence
  • Our experience
  • Uncertainty

What do you consider when making a decision?

This can be represented like so:

Getting quantitative

This framework is similar if you are approaching the decision quantitatively (i.e. using models and data).

Using models and data when making a decision

A hypothetical example

  • data (points) are the evidence
  • experience (current state of knowledge) is used to specify the model (e.g. linear)

The relationship between effort and reward is nearly 1 to 1, suggesting that the more effort you invest, the more reward. That said, there is scatter around in the points around the 1:1 line, suggesting uncertainty.

Iterative decision-making

  • Few decisions are once-off
  • Evaluating the outcome is crucial, so we can learn from experience
    • Was your forecast was any good?
    • Should you refine or replace your model, consider additional scenarios or inputs?

Our hypothetical example

Revisiting our Effort to Reward example, what would you do if the decision-maker decided to invest huge effort, but the next few data points looked like this?

Iterative decision-making and the scientific method

The iterative decision making cycle mirrors the scientific method, i.e.:

Observation > Hypothesis > Experiment > Analyze > Interpret > Report > (Repeat)

So iterative decision-making facilitates iterative learning (i.e. scientific progress).

The importance of prediction in ecology

“prediction is the only way to demonstrate scientific understanding” - Houlahan et al. 2017

If we cannot make reasonably good predictions, we’re missing something.

In ecology, we mostly test qualitative, imprecise hypotheses:

  • “Does X have an effect on Y?”, rather than:
  • “What is the relationship between X and Y?” or better yet - “What value would we expect Y to be, given a particular value of X?”

Without testing precise hypotheses and using the results to make testable predictions we don’t know if our findings are generalizable beyond our specific data set.

  • If our results are not generalisable, then we’re not making progress towards a better understanding of ecology.

Iterative near-term ecological forecasting


Seeks to make prediction a central focus in ecology, on a time scale that is both useful for decision makers and allows us to learn from testing our predictions (i.e. days to decades)


  • The “gold standard” is an informatics pipeline that can ingest new data and make new forecasts automatically with minimal user input.
  • This poses a number of major challenges and requires a big improvement in quantitative skills in biology (hence this course…).
  • Fortunately, any steps towards the gold standard is likely to be useful, even if you never get there.

Iterative near-term ecological forecasting

Step 1: Start with your initial conditions (data and knowledge that feed into designing and fitting your model)

Iterative near-term ecological forecasting

Step 2: Make forecasts (i.e. predictions into the future - in blue) using your model, based on your initial conditions (red).

Iterative near-term ecological forecasting

Step 3: Monitor and collect new observations (green) to compare with your forecasts (blue) and original observations (i.e. initial conditions (red)).

Iterative near-term ecological forecasting

Step 4: Analyze the new observations in the context of your forecasts and original observations, and update the initial conditions for the next iteration of the forecast.

Iterative near-term ecological forecasting

This can also be represented as a cycle, mirroring the scientific method:

The key steps are:

  1. Make a forecast based on initial conditions
  2. Compare your forecast to new observations
  3. Analyze the new observations in the context of your forecast and original data
  4. Update estimates of the current state of the system (data and understanding), before making a new forecast

Iterative near-term ecological forecasting

Two things not indicated in this diagram are:

  • comparing forecasts and new observations allows you to learn about the various sources and drivers of uncertainty in your forecast
    • this can be used to refine your model or adapt or guide what and how to monitor so that you can reduce those uncertainties
  • iterative ecological forecsts require automated informatics pipeline, which are best done in a reproducible research framework

Iterative near-term ecological forecasting

Iterative ecological forecasts are thus aimed at:

  1. applied outcomes, through providing evidence to support decision making
  2. knowledge generation through iterative learning i.e. the scientific method

So it’s a great way of getting scientists to engage in real-world problems, demonstrating the value of our science, and learning by doing!

Iterative ecological forecasting in context

This figure from Dietze et al. 2018 provides an expanded representation of these conceptual links between iterative ecological forecasting, the scientific method, and decision making (here in the context of adaptive management, which is a management paradigm that focuses on learning by doing).

Reproducible research

Iterative ecological forecasts need to be founded on a highly efficient informatics pipelines that are robust and rapidly updateable.

  • The emphasis is on near-term forecasts to inform management. If the process of adding new data and updating forecasts is too slow, the value of the forecasts is lost.

  • The best way to build the ecoinformatics pipeline is to follow reproducible research principles, including good (and rapid) data management

Adding this to previous slide highlights what I like to think of as “The Olympian Challenge of data-driven ecological decision making”.